Automatic segmentation of thin sections with GRASS GIS: first experiments

Micromorphological analyses of soil thin sections from archaeological sites are fundamental tools in geoarchaeology, which includes the study of site formation processes. The thin sections are observed under the optical microscope in polarized light with different lighting conditions (mainly with plane polarized light and with crossed nicols) and digitized with high resolution. The joint analysis of the resulting images allows the detection of significant features for the archaeological interpretation of the site. This work describes some experiments for the automatic or semi-automatic analysis of this kind of images using modules for image segmentation and filtering available in GRASS GIS. Single images are segmented using the different approaches, variational and region growing, implemented in GRASS r.smooth.seg and i.segmentI modules. The resulting images are filtered using threshold values evaluated for each image. Finally, the filtered images with different types of polarization are combined. The resulting map is checked against a reference map created manually by digitizing the contour of relevant areas. This work presents the first experiments for the optimal choice of the segmentation parameters for the variational approach and the threshold values using automatic or semi-automatic approaches.